Alex Savanovich

In IT since 2004, Alex is one of ScienceSoft’s leading specialists with over seven years of experience in data science and AI.

With deep expertise in Python, C++, and advanced math, Alex drives the creation of unique machine learning, computer vision, and natural language processing (including LLMs) algorithms that reach 98% accuracy. He also actively participates in the development of solutions based on GPT- and BERT-like models. Alex led the development of proprietary algorithms for ScienceSoft’s most prominent AI projects, including drilling equipment monitoring software and a clay pigeon shooting tracking solution.

Alex’s responsibilities go beyond mere ML/AI model training. As a senior data scientist, he takes part in direct communication with ScienceSoft’s clients so that both sides get a clear understanding of how the business requirements can be met within the existing technology constraints.

Designing a feasible AI solution for a unique project can be more difficult than just training a standalone ML model. More often than not, we have to combine several algorithms and models at different data pipeline stages, since you can only reach about 80% accuracy with standard tools. 90% is also doable if you really put your mind to it, but anything close to 99% requires custom solutions and non-trivial approaches. You should account for the existing systems that AI will be integrated with, the real-world data that the algorithm should adapt to, and the way users will interact with the output. Such a solution-focused approach requires a thorough investigation of the client’s goals and IT environment. That way, we can deliver value-driving solutions rather than just unaligned intelligence.